ORIGINAL ARTICLE

Measurement of circulating transcripts and gene cluster analysispredicts and defines therapeutic efficacy of peptide receptorradionuclide therapy (PRRT) in neuroendocrine tumors

L. Bodei1,2 & M. Kidd3& I. M. Modlin2,4

& S. Severi5 & I. Drozdov6 & S. Nicolini5 &

D. J. Kwekkeboom2,7& E. P. Krenning2,7 & R. P. Baum2,8

& G. Paganelli5

Received: 3 August 2015 /Accepted: 30 October 2015 /Published online: 23 November 2015# Springer-Verlag Berlin Heidelberg 2015

AbstractBackground Peptide receptor radionuclide therapy (PRRT) isan effective method for treating neuroendocrine tumors(NETs). It is limited, however, in the prediction of individualtumor response and the precise and early identification ofchanges in tumor size. Currently, response prediction is basedon somatostatin receptor expression and efficacy by morpho-logical imaging and/or chromogranin A (CgA) measurement.The aim of this study was to assess the accuracy of circulatingNET transcripts as a measure of PRRTefficacy, and moreoverto identify prognostic gene clusters in pretreatment blood thatcould be interpolated with relevant clinical features in order todefine a biological index for the tumor and a predictive quo-tient for PRRT efficacy.Methods NET patients (n=54), M: F 37:17, median age 66,bronchial: n=13, GEP-NET: n=35, CUP: n=6 were treatedwith 177Lu-based-PRRT (cumulative activity: 6.5-27.8 GBq,median 18.5). At baseline: 47/54 low-grade (G1/G2; bronchi-al typical/atypical), 31/49 18FDG positive and 39/54 progres-sive. Disease status was assessed by RECIST1.1. Transcripts

were measured by real-time quantitative reverse transcriptionPCR (qRT-PCR) and multianalyte algorithmic analysis(NETest); CgA by enzyme-linked immunosorbent assay(ELISA). Gene cluster (GC) derivations: regulatory network,protein:protein interactome analyses. Statistical analyses: chi-square, non-parametricmeasurements, multiple regression, re-ceiver operating characteristic and Kaplan–Meier survival.Results The disease control rate was 72 %. Median PFS wasnot achieved (follow-up: 1–33 months, median: 16). Onlygrading was associated with response (p<0.01). At baseline,94 % of patients were NETest-positive, while CgA was ele-vated in 59 %. NETest accurately (89 %, χ2=27.4; p=1.2×10−7) correlated with treatment response, while CgAwas 24%accurate. Gene cluster expression (growth-factor signalomeand metabolome) had an AUC of 0.74±0.08 (z-statistic=2.92, p<0.004) for predicting response (76 % accuracy).Combination with grading reached an AUC: 0.90±0.07, irre-spective of tumor origin. Circulating transcripts correlated ac-curately (94 %) with PRRT responders (SD+PR+CR; 97 %)vs. non-responders (91 %).

L. Bodei and M. Kidd contributed equally to this work.

* I. M. Modlinimodlin@optonline.net

1 Division of Nuclear Medicine, European Institute of Oncology,Milan, Italy

2 LuGenIum Consortium, Milan, Rotterdam, Bad Berka,London, Italy, Netherlands, Germany, UK

3 Wren Laboratories, Branford, CT, USA4 Yale School of Medicine, 310 Cedar St, New Haven, New

Haven 06510, CT, USA

Eur J Nucl Med Mol Imaging (2016) 43:839–851DOI 10.1007/s00259-015-3250-z

5 Nuclear Medicine and Radiometabolic Units, Istituto ScientificoRomagnolo per lo Studio e la Cura dei Tumori (IRST) IRCCS,Meldola, Italy

6 Bering Limited, London, UK

7 Nuclear Medicine Department, Erasmus Medical Center,Rotterdam, The Netherlands

8 Theranostics Center for Molecular Radiotherapy and Imaging,Zentralklinik Bad Berka, Bad Berka, Germany

Conclusions Blood NET transcript levels and the predictivequotient (circulating gene clusters+grading) accurately pre-dicted PRRT efficacy. CgAwas non-informative.

Keywords Neuroendocrine tumor . Chromogranin .68Ga-PET . Gene transcripts . NETest . PRRT

Introduction

Neuroendocrine tumors (NETs) are frequently identified whenthey are metastatic or locally advanced. Their pathobiologysupports multiple treatment modalities, which are individual-ized through a multidisciplinary approach. In principle, thechoice of therapy depends on individual tumor characteristicsand ranges from complete surgical eradication to a Bwatch andwait^ approach [1].

Peptide receptor radionuclide therapy (PRRT) with 177Lu-DOTA-Tyr3-Thr8-octreotide (177Lu-octreotate) is anestablished effective therapeutic modality that has been usedfor 15 years in the treatment of unresectable or metastaticgastroenteropancreatic (GEP) and bronchopulmonary (BP)NETs. Objective response rates of 15–35% are common, withmodest toxicity in the majority of cases if the necessary pre-cautions (e.g., renal protection) are undertaken [2]. Of partic-ular significance is outcome; both progression-free survival(PFS) and overall survival compare favorably with somato-statin analogues (SSA), chemotherapy and new Btargeted^therapies [3].

Combinations of anatomical (CT/MRI) and functional im-aging (OctreoScan® or 68Ga-SSA-PET/CT–SRI) techniquesare used to monitor therapy [4]. However, the low-dimensional alterations typical of these tumors, coupled withthe resolution limits of both CT and PET scanners, limit theearly and accurate detection of lesion alterations, particularlyfor large-volume, multicentric or poorly demarcated disease,rendering RECISTcriteria largely inadequate for capturing theeffects of therapy in a timely and accurate fashion [4].

A variety of blood markers, particularly chromogranin A(CgA), have been proposed to facilitate the determination oftherapeutic efficacy and disease status. None, however, hasshown adequate sensitivity and specificity or met the metricstandards of accuracy [5–7]. Nevertheless, despite reserva-tions regarding its efficacy, CgA remains the default measure-ment [8]. Given the key unmet need for sensitive and accuratebiomarkers to define therapeutic efficacy, innovative solutionsincluding circulating tumor cell (CTC) analysis [9], miRNAmeasurement [10] and multigene assays with algorithmicanalyses (MAAA) have been proposed [11, 12]. The last ofthese assesses NET biological activity using gene inferencetechnology and cancer hallmark prediction [12]. The set ofcirculating transcripts that defines this Bfingerprint^ exhibitshigh sensitivity (98 %) and specificity (97 %) for detecting

NETs, is standardized and reproducible (inter- and intra-assayCV <2 %) and outperforms other GEP-NETs biomarkers, in-cluding CgA [11].

Blood gene transcript analysis in patients treated with SSAtherapy or surgery has demonstrated a significant advantage inthe early detection of residual disease [13] and in the assess-ment of somatostatin analog response [14]. Furthermore, com-binations of circulating NET transcripts in conjunction with68Ga-SSA-PET/CT have demonstrated that molecular imag-ing parameters (SUVmax) could be integrated into a predictivequotient of tumor status [15]. We recently demonstrated thatthe segregation of circulating NET transcripts into gene clus-ters using unbiased protein:protein interactome approaches, inaddition to regulatory network and evaluation of publisheddata, was successful in defining the NET cancer Bhallmarks^[12, 16]. With this strategy, we identified nine genomic clus-ters, or Bomes,^ governing the various cellular functions(SSTRome, proliferome, growth factor signalome, metabo-lome, secretome, epigenome, plurome, apoptome) that cap-tured the biological activity of the tumor. Inclusion of theseBomes^ provided a predictive activity index for defining tu-mor behavior and outcome [12]. A recent consensus publica-tion considered gene transcript analysis to be the biomarkerstrategy most likely to provide clinical utility [17].

The purpose of the present study was to evaluate the effi-cacy of PRRT with 177Lu-octreotate using circulating NETtranscripts and CgA. In addition, we sought to identify wheth-er circulating NET transcripts or gene clusters in the pre-treatment blood, which provide biologically relevant informa-tion on the individual tumor, could be interpolated with rele-vant clinical features and used to predict the response toPRRT.

Materials and methods

Patients

Patients with GEP and BP NETs (n=78) that exhibited so-matostatin receptors (identified by receptor imaging) andwho were candidates for PRRT with 177Lu-octreotate wereenrolled from July 2012 to April 2015. The present study isan interim analysis of 54 individuals with complete follow-up(at least 6 months post-therapy) as of April 2015.

Study design and procedures

PRRT protocol Patients were treated according to two proto-cols, depending on previous treatments and risk factors fordelayed toxicity [18, 19] (Table 1). The first protocol wasdeveloped for PRRT-naïve patients. Two different levels ofintended cumulative therapeutic activity, 18.5 (1A) or 27.8(1B) GBq divided into four cycles (4.6 and 6.5 GBq each,

840 Eur J Nucl Med Mol Imaging (2016) 43:839–851

respectively), were administered 2 months apart based uponkidney function and bone marrow reserve. The second proto-col (14.8 GBq divided into four cycles of 3.7 GBq each, ad-ministered 2 months apart) was utilized in individuals pre-treated with PRRT (177Lu-octreotate or 90Y-octreotide) [20].All participants provided informed consent for PRRT andtranslational analysis, which was authorized by the ethicscommittees (PRRT: IRST 100.06, EudraCT: 2011-002891-18, 04/08/2011; NETest: IRST B007 [70/12], 10/10//2012).

Assessment of therapeutic response CT (or MRI) was per-formed at baseline (within 3 months from the start of PRRT)and 3 and 6 months after PRRT to define disease status andassess response according to RECIST 1.1 criteria (Table 2)[21]. Baseline status was defined according to RECISTcriteria, based on a comparison with CT/MRI obtained within1 year from enrolment. 68Ga-SSA-PET or OctreoScan® wasperformed at baseline and at 6-month follow-up. Subsequentimaging was performed per standard clinical practice.

Transcript analysis Samples of 10 ml of whole blood werecollected in 2×5ml tubes and snap-frozen at baseline and aftereach administration of 177Lu-octreotate, and at 3- and 6-monthfollow-up after treatment completion (Table 2). Plasma CgAwere collected at the same time points. A two-step protocol(RNA isolation, cDNA production and PCR) was used [11,12]. Analyses were carried out using the MATLAB Statistics

and Machine Learning Toolbox (MATLAB R2011a;MathWorks, Natick, MA, USA). The NETest mathematicallydelineates disease activity risk on a scale of 0 to 100 %, asfollows: minimal, <14 %; low, 14–47 %; high, >47 %. TheNETest also includes biologically relevant gene expressionmeasurements from the different Bomes^ (SSTRome,proliferome, metabolome, secretome, epigenome andplurome) that differentiate progressive from stable disease[12].

CgA assay CgA was measured using NEOLISA™Chromogranin A kit (Euro Diagnostica AB, Malmö,Sweden). The upper limit of normal was 108 ng/ml [11].

Toxicity evaluation Hematological, renal, and liver toxicitywas evaluated using Common Terminology Criteria forAdverse Events (CTCAE) Version 4.0 (NCI, Bethesda, MD,USA).

Statistical analysis Sensitivity comparisons between theNETest and CgA were conducted using chi-square, non-parametric measurements and receiver operator curve (ROC)analysis with determination of the area under the curve(AUC). Prism 6.0 for Windows (GraphPad Software, LaJolla CA USA, www.graphpad.com) and MedCalcStatistical Software version 12.7.7 (MedCalc Software bvba,Ostend, Belgium; http://www.medcalc.org; 2013) wereutilized. Multiple regression analyses were undertaken toidentify significant clinical parameters, e.g.,18FDG-positivity. As reported in the literature, pre-therapy CgA>600 ng/ml was used as one measure to predict response andPFS [3, 22].

An BOme Index^ was derived from a summation of genecluster expression that included the Bgrowth factor (GF)signalome^ and the Bmetabolome^ as measured in the pre-treatment blood. These were chosen based on expressionlevels and AUC values >0.65. Ome index values were sepa-rated into two groups based on their ability to predict re-sponse. On the basis of >85 % specificity for predicting dis-ease response (see "Results"), cut-off values of 5.9 separatedBOme (high)^ (>5.9) from BOme (low)^ (<5.9). A "PredictionQuotient" comprising the grade (low grade=G1/G2, well-dif-ferentiated, bronchial typical or atypical carcinoid^; highgrade=G3, Bpoorly differentiated^) and Ome Index (see

Table 1 PRRT protocols. Rationale for choosing the protocol was based on PRRT naiveté and the presence of Brisk factors^ for delayed toxicity.Cycles were administered 2 months apart

Status prior to PRRT PRRT protocol Intended cumulative activity (GBq) Intended activity per cycle (GBq) Patients treated

PRRT naïve, risk factors 1A 18.5 in 4 cycles 4.6 22

PRRT naïve, no risk factors 1B 27.8 in 4 cycles 6.5 17

PRRT pre-treated 2 14.8 in 4 cycles 3.7 15

Table 2 Schedule of blood sampling and diagnostic exams(morphologic and functional)

Assessment

1st cycle Blood for transcript analysis and CgACT/MRISRI*

2nd cycle Blood for transcript analysis and CgA

3rd cycle Blood for transcript analysis and CgA

4th cycle Blood for transcript analysis and CgA

3-month follow-up Blood for transcript analysis and CgA,CT/MRI

6-month follow-up Blood for transcript analysis and CgA,CT/MRISRI*

*SRI: 68 Ga-SSA-PET or OctreoScan

Eur J Nucl Med Mol Imaging (2016) 43:839–851 841

"Results") was developed. In order to add further specificity tothe Prediction Quotient, we separated the components intotwo groups—Low Grade/High Ome and High Grade/LowOme—to assess clinical utility (prediction of disease responseand PFS). The mathematical indices evaluated included theNETest (algorithm of 9 Bomes^ and 51 NET genes), theOme index (GF signalome+metabolome) and the PredictionQuotient (comprising Ome index+grade). The predictive ac-curacy of each of the mathematical indices (Fig. 1) was com-pared using ROC curve analyses, and the sensitivity, specific-ity, and the AUCwere calculated (MedCalc) [23]. AUCs werecompared and the Z-statistic derived [24] (MedCalc). Kaplan-Meier survival curves (PFS) were generated and analyzed inPrism. Log-rank (Mantel–Cox) and hazard ratios (Mantel–Haenszel) were calculated. Z-statistic scores >1.96 are signif-icant (p<0.05). For other metrics, e.g., accuracy, 80 % is gen-erally accepted as the acceptable cut-off. [6, 7] All data wereexpressed as mean±standard error of mean.

Results

A. Clinical audit As of April 2015, 78 GEP and BP NETpatients were consecutively enrolled, 60 of whom had defin-itive data. Six were excluded because of withdrawal of con-sent (three subjects), death during PRRT for causes other thanNET (two subjects) or missing data before dropping out due toprogression (one subject). The remaining 54 were included inthe analysis. Follow-up ranged from 1 to 33 months; the me-dian was 16 months (Table 3).

B. Treatment response and PFS Cumulative administeredactivity of 177Lu-octreotate was 6.5–27.8 GBq (median 18.5).Fifty-one of the 54 patients (94.4 %) received ≥75 % of theintended cumulative activity. PRRT was well-tolerated, with-out serious side effects. Severe hematological toxicity waslimited to a minority of patients (8 % G3; no G4), whilemoderate (G2; no G3/4) creatinine toxicity occurred in 5 %.Response to PRRT included complete response (CR) in one

subject (1.9 %), partial response (PR) in nine (16.7 %), stabledisease (SD) in 29 (53.7 %), and progression (PD) in 15(27.8 %), with a disease control rate (CR+PR+SD) of72.2 % (no significant differences between the two PRRTprotocols). At the time of the analysis, the median PFS hadnot been not reached (Fig. 2a). While no significant differ-ences were noted by site (p=0.11) (Fig. 2b), the highest ob-jective response rate occurred in pancreatic NETs.

Regression analysis of clinical parameters at baseline(demographics, SSA use, primary location, metabolic sta-tus at18FDG PET/CT, symptoms, time since diagnosis,baseline ECOG status) identified that no clinical parame-ters were associated with treatment response (odds ratios:0.5 [SSA, p=0.35)] – 2.0 [ECOG status, p=0.35]) exceptfor grading (OR: 8.75, p=0.004) (Table 4). Further analy-sis confirmed that grading was associated with outcome(R2=0.19, F-ratio: 11.7, p=0.0012).

C. Biomarker assessment NETest and CgA were examinedas a function of cycle and follow-up time in order to eval-uate the relationship to treatment response (Table 2).Outcome was defined by RECIST 1.1 criteria as responders(SD+PR+CR; n=39) or non-responders (PD; n=15).

C1. NETest The NETest (mean pre-therapy time 0, prior toCycle I) was 42.9±4.6 %. This was elevated (>14 %) in 94 %(Fig. 3a). At 3-month follow-up, this was reduced (33.2±5.6 %, p=0.08), and at the 6-month follow-up, the reductionwas significant (25±3.7 %, p=0.012) in responders. At 3 and6 months, levels were increased in non-responders (3 months:45.3±8 %, p=0.32 vs. pre-therapy, p=0.11 vs. responders and6 months: 58.2±6.2 %, p=0.04 vs. pre-therapy, p<0.0001 vs.responders). The NETest decreased in 88 % of responders (nochange in 12 %) and was increased in 90 % (no change in10 %) of non-responders at 6 months.

C2. CgA Levels prior to therapy were 2426±611 ng/ml(Fig. 3b). Elevated (>108 ng/ml) CgA, however, was onlynoted in 57 %. The decrease (to 689±215 ng/ml) in re-sponders (at 6 months) was not significant. CgA was un-changed in 47 %, decreased in 21 % and elevated in 32 %of responders at 6 months. Levels were significantly de-creased in non-responders (202±117 ng/ml) (p=0.045).CgA was unchanged in 33 %, decreased in 40 % and waselevated in 27 % of the non-responders. No significant differ-ences were noted at the 3-month or 6-month follow-up timepoints (p>0.3).

C3. NETest vs CgAThe NETest correlated significantly withpre-therapy status (94 % vs. 57 %, p=0.0004, chi-square)(Fig. 3c) in comparison to CgA, while changes in levelsmore consistently correlated with treatment response thanin CgA (responders: p=0.0002, non-responders: p=0.0068)

Fig. 1 Development of the Predictive Quotient for assessing PRRTtherapy. Graphic demonstrating the development of the PredictionQuotient from an assessment of circulating biomarkers as well asclinically relevant information

842 Eur J Nucl Med Mol Imaging (2016) 43:839–851

(Fig. 3d–e). Neither baseline (pre-treatment levels) of theNETest nor of CgA were predictive of outcome (NETest:OR=0.98, p=0.3; CgA: OR=1.0, p=0.18).

C4 Metrics The metrics for biomarkers and outcomeidentified that the NETest had 89 % accuracy, 75 %sensitivity, 100 % specificity, 100 % PPV and 83 %NPV (Fig. 4a). CgA, in contrast, had 24 % accuracy,17 % sensitivity, 40 % specificity, 40 % PPV and 17 %NPV (Fig. 4b). The NETest significantly outperformedCgA (chi-square=27.4; p=1.2×10−7).

D. Derivation of outcome predictors We next examinedeach of the biomarkers or constituent factors, e.g., gene clus-ters or Bomes^ and clinical parameters or combinations there-of, to predict response. We specifically focused on pre-therapyblood.

D1. Biomarkers Pre-therapy (T0 cycle I) NETest or CgAlevels alone were not significantly predictive of outcome(see Section C3). ROC analysis identified AUCs of 0.65±0.09 (p=0.09) and 0.58±0.08 (p=0.32), respectively. CgAlevels (>600 ng/ml) were also not associated with outcomeor PFS (<600 ng/ml: 16.9±1.2 months PFS vs. 13.8±.4 months, p=0.19). There was no significant difference inNETest (p=0.26-0.71) or CgA (p=0.06-0.6) levels betweentumor sites, e.g., bronchopulmonary versus GEP-NET.

Table 3 Patient Demographics and Disease Characteristics

Patients (n=54)

Age, median (range) in years 66 (43–83)

Gender, n 37 M, 17 F

Time since diagnosis, in months (range) 3–265

Mean (SD) 58 (57.5)

Median 37

Length of follow up, median (range), in months 16 (1–33)

NET origin, n (%)

Broncho-pulmonary [26] 13 (24.1 %)

Typical carcinoids 1 (1.9 %)

Atypical carcinoids 7 (13 %)

NOS carcinoids 1 (1.9 %)

High-grade 4 (7.4 %)

GEP 35 (64.8 %)

Stomach 1 (1.9 %)

Pancreas 14 (25.9 %)

Small intestine 17 (31.5 %)

Colon 1 (1.9 %)

Rectum 2 (3.7 %)

Unknown 6 (11.1 %)

GEP NETs, Tumor grade (WHO 2010 [25]), n (%)

G1 (Ki-67 0–2 %) 6 (17.1 %)

G2 (Ki-67 3–20 %) 20 (57.1 %)

G3 (Ki-67>20 %) 3 (8.6 %)

Non-specified (well-differentiated) 6 (17.1 %)

Initial clinical stage, n (%)

Stage IV 54 (100 %)

Liver 45

Lymph nodes 31

Bone 18

Peritoneum 9

Lung 4

Other sites (e.g. adrenal, pleura, pericardium) 9

Baseline tumor status

Progressive disease 39 (72.2 %)

Stable disease 13 (24.1 %)

Response to previous chemotherapy 2 (3.7 %)

Extent of disease§, n (%)

Limited 7 (13 %)

Moderate 26 (48.1 %)

Extensive 21 (38.9 %)

Intensity of uptake*, n (%)

Grade 1 1 (1.9 %)

Grade 2 6 (11.1 %)

Grade 3 16 (29.6 %)

Grade 4 31 (57.4 %)18FDG PET/CT

Negative 18 (33.3 %)

Positive 31 (57.4 %)

Not tested 5 (9.3 %)

Table 3 (continued)

Patients (n=54)

CgA

Normal 23 (42.6 %)

Elevated 31 (57.4 %)

Previous therapy, n (%)

Surgery 32

Primary tumor surgery 30

Liver surgery 7

Non-resective surgery 1

Somatostatin Analogs 44

Pharmacotherapy

Chemotherapy 21

Everolimus 5

Sunitinib 1

Interferon alpha 1

Other

PRRT 16

Radiotherapy 6

TACE 4

§Extension of uptake according to Krenning scale [32]

*Intensity of uptake according to Krenning scale either on OctreoScan[32] or modified criteria applied to 68 Ga-PET

Eur J Nucl Med Mol Imaging (2016) 43:839–851 843

D2: BOmes^ An assessment of gene expression in the ninedifferent Bomes^ captured by the NETest identified no sig-nificant difference between primary sites (p=0.10–0.98).Pre-treatment blood levels of growth factor signaling (GFsignalome) and metabolism, however, were different be-tween responders and non-responders. Specifically, subse-quent PRRT responders exhibited significantly elevatedgrowth factor signaling (9.4±1.3 vs. 5.3±0.7, p=0.05) andmetabolomic gene expression (4.37 vs. 2.3±0.6, p=0.03) at

T0 (Cycle I, i.e., prior to therapy) compared to non-responders (Fig. 5a). Gene expression of the other Bomes^,e.g., somatostatin receptor-ome BSSTRome^, were not dif-ferent (49.5±11.6 vs. 29.7±8.5, p=0.15). ROC analysis(Fig. 5b) identified AUCs of 0.67 and 0.7 and z-statisticsof 2.44 and 2.26 for the GF signalome and metabolome,respectively. An integration of the two Bclusters^ (GFsignalome+metabolome) in an BOme Index^ resulted in anAUC of 0.74±0.08 (z-statistic=2.915, p=0.0036). A cut-offof 5.9 (normalized gene expression) exhibited >85 % spec-ificity for predicting response (>5.9 predicted subsequentPRRT responders).

D3. Clinical parameters Multiple regression analysis ofclinical parameters at baseline identified that only grading[low grade: G1/G2 (WHO 2010 [25]), bronchial typical oratypical carcinoids [26]]; high grade: G3, poorly differenti-ated) was associated with outcome (coefficient: 0.598±0.19,p=0.0035). Individually, the coefficient for G1 was −0.71(p=0.004) and for G 2 was −0.66 (p=0.0043). Variablesincluding syndrome, SRI parameters (extension of disease,intensity of uptake [Rotterdam Scale proportions] pre-PRRT), concurrent SSA use, baseline status or syndromewere not significantly associated with PRRT response in thiscohort (Tables 4 and 5).18FDG positivity, although not asso-ciated with outcome (OR: 1.24, p=0.74), was associatedwith PFS (chi-square=15.9, p<0.0001, log rank: 5.1).Low-grade tumors typically responded to therapy (77 %),while 50 % of high-grade lesions exhibited a response(Fig. 5c). While grade per se, was not significantly different(p=0.12), grading did exhibit an AUC of 0.66±0.06 (p=0.08) for predicting response (Fig. 5d).

E. Derivation of a clinically useful BPrediction Quotient^The combination of the Ome Index (BGF signalome^+Bmetabolome^, Fig. 5b) with the grade had a significantly

Fig. 2 PRRT outcome by site assessed by overall RECIST objectiveresponse. a) Progression-free survival curve (solid red line) with 95 %confidence interval (dotted red lines) indicates that a median value wasno t r eached . b ) . Responde r s (SD+PR+CR) were 62 %

bronchopulmonary, 71 % GI, 93 % pancreas and 67 % unknown. Chi-square analysis identified no significant differences (p=0.11) in outcomebased on site. BP, bronchopulmonary (n=13); GI, gastrointestinal (n=20); PANC, pancreas (n=15), UNK=unknown site (n=6)

Table 4 Clinical Variables and Response*

Variable Logistic Regression Analysis MultipleRegressionAnalysisOR (95 % CI) p Value

Age 1.01 (0.94–1.078) 0.85

Gender 0.89 (0.25–3.17) 0.86

Site

Bronchopulmonary 0.73 (0.17–3.1) 0.67

Pancreas 0.64 (0.15–2.7) 0.54

Gastrointestinal tract 2.29 (0.68–7.69) 0.18

Unknown 0.49 (0.05–4.54) 0.53

Time since diagnosis 1.01 (0.997–1.02) 0.17

Grading 8.75 (2.01–38.14) 0.004 R2=0.19,F-ratio 11.7,p=0.0012

Baseline status 1.78 (0.423–7.48) 0.43

ECOG 2.00 (0.475–8.42) 0.35

Syndrome 1.56 (0.47–5.22) 0.47

Extent of disease 1.57 (0.62–3.96) 0.34

Intensity of uptake 0.81 (0.38–1.73) 0.58

FDG 1.24 (0.35–4.44) 0.74

SSA use 0.5 (0.12–2.11) 0.35

*RECIST 1.1 Criteria

844 Eur J Nucl Med Mol Imaging (2016) 43:839–851

better AUC (0.90±0.06) than the grade alone (AUC=0.66,difference between areas 0.23, z-statistic 2.25, p=0.024)(Fig. 6a) for predicting response. This Prediction Quotientwas also clinically useful. The high-grade/low-Ome tumorgroup had significantly lower PFS (17 months) than the low-grade/high-Ome group (undetermined, log-rank: 26.8;p<0.0001: Fig. 6b). The hazard ratio was 53.26. A separateanalysis of both BP-NETs and GEP-NETs showed that thePrediction Quotient exhibited an AUC=1 (vs. 0.94 for theOme Index and Grade: 0.75) for the former while this was0.88 (vs. 0.85 for the Ome Index and Grade: 0.63) for GEP-NETs. The overall metrics for the Prediction Quotient were:accuracy of 94 %, sensitivity 79 %, specificity 100 %, PPV100 % and NPV 93 % (Fig. 6c).

Discussion

Our study was designed to assess whether a circulatingmultianalyte 51-gene NET signature could be used as a surro-gate measure of clinical responses to PRRT when assessed at3- and 6-month follow-up.We also evaluated whether integra-tion of the NET gene signature (gene cluster analysis) mea-sured in pre-treatment blood and clinical factors (e.g., tumorgrading, pre-PRRT FDG positivity or pre-PRRT SRI uptake)could be used to predict outcome.

Overall, PRRT was effective (72 % disease control rate)and well-tolerated in advanced metastatic NETs, even whenrisk factors for delayed renal or hematological toxicity werepresent, including individuals previously treated with PRRT

Fig. 3 NETest and CgA levels in responders and non-responders. a)NETest was significantly reduced at 6-month follow-up compared topre-treatment values. Non-responders (NR) exhibited significantly elevat-ed levels at 6 months (FUP_6m). This was also significantly higher thanin responders (R). b) A significant alteration was noted only for CgA innon-responders (NR) at 6 months. c) Pre-PRRT: NETest scores wereelevated in 94 % and CgA in 57 % (p=0.0004). d) Responders: NETestdecreased in 88 % of responders (no change in 12 %). CgA was

unchanged in 47 %, decreased in 21 % and was elevated in 32 % (p=0.0002 vs. NETest). e) Non-Responders: NETest increased in 90 % (nochange in 10 %). CgA was unchanged in 33 %, decreased in 40 % andwas elevated in 27 % (p=0.0068 vs. NETest). For the NETest, no falselydecreased or increased values occurred in responders or non-responders,respectively. Mean±SEM. *p<0.05 vs. Cycle I (T0); #p<0.05 vs. re-sponders (6-month follow-up)

Eur J Nucl Med Mol Imaging (2016) 43:839–851 845

[20, 27]. Stability was included in the positive outcome, giventhe intrinsic malignant nature of these tumors. Responses oc-curred within 2–8 months (median 3), consistent with previ-ous observations [28, 29]. Median PFS was not reached, andPFS was significantly longer in low-grade (undefined vs.10 months in high-grade lesions) and in18FDG-negative (un-defined vs. 15 months in FDG-positive) tumors.

Of the potential clinical parameters, tumor grading alonehad some value in predicting GEP- and BP-NETs most likelyto respond to PRRT. [30, 31] The metabolic status (18FDG-PET) predicted PFS but did not reach statistical significance atmultivariate analysis in our cohort. This may be due to the lownumber of patients with negative18FDG-PET at baseline (18vs. 31). Neither clinical syndrome nor pre-PRRT SRI uptakewas associated with outcome. The latter is in contrast to aprevious report [32]. The non-homogeneous patient popula-tion, different PRRT dosages and receptor techniques used(scintigraphy, PET) are possible confounders.

With regard to circulating biomarkers and treatmentresponse, changes in NETest accurately (89 %) correlat-ed with treatment response, while CgA was only 24 %accurate. In two previous studies, pharmacological treat-ment with SSAs or surgery [11, 13], we demonstrated ahigh correlation between the NETest and tumor re-sponse, indicating that this circulating fingerprint effec-tively captures the tumor response to intervention. Toprovide further insight into the prediction of treatmentefficacy and assessment of response, we developed agene inference methodology to identify specific Bomics^(growth factor signalome and tumor metabolome) rele-vant to tumor biology (behavior). We integrated theseinto a predictive quotient. We focused on the GFsignalome and metabolome, as they both provide bio-logically relevant information about the tumor [12]. The

former (including ARAF1, BRAF, KRAS and RAF1) cap-tures information specific to growth factor-mediated sig-naling. Growth factor expression and signaling is awell-known component of GEP-NET proliferation, andsignaling pathways (e.g., the RAS/RAF/MAPK signalingpathway) are typically activated [33, 34]. Moreover, B-Raf expression can be detected by immunohistochemis-try (in 75 % of tumors) [35], while RAF1 is directlylinked to proliferation [36]. Genes in the metabolome(ATP6V1H, OAZ2, PANK2, PLD3) are not as well-characterized in GEP-NETs, but their biological rolesare well-described. ATP6VIH is involved in regulatingneuroendocrine oxidative phosphorylation (pancreatic is-lets) [37], OAZ2 is involved in polyamine biosynthesis[38], PANK2 in metabolism and oxidation [39], andPLD3 in lipid metabolism and hypoxic signaling [40].

Expression of signaling and metabolic genes in bloodprior to initiation of PRRT showed an AUC of 0.74±0.08(z-statistic=2.92, p<0.004) for predicting subsequent treat-ment response (76 % accuracy). When combined withgrading, the accuracy increased significantly (76 to94 %) and was highly effective as a prognostic marker(AUC: 0.90±0.07), irrespective of tumor origin. Tumorsthat exhibited significant signalome/metabolome geneelevatins at baseline were responsive (97 %) to PRRT ascompared to those who failed therapy. Combining thesetwo clusters in an BOme Index^ and then adding theseinto a Predictive Quotient with grading provided signifi-cantly greater predictive accuracy than that provided bythe original parameters. In comparison, CgA (amonoanalyte constitutive secretory protein) was non-infor-mative, and the variations after the completion of PRRTwere non-significant and failed to correlate with objectiveresponse. Of interest, however, was that CgA was lowest

Fig. 4 Metrics for the NETestand CgA in responders and non-responders. a) The NETest:accuracy 89 %, sensitivity 75 %,specificity 100 %, PPV 100 %and NPV 83%. b) CgA: accuracy24%, sensitivity 17%, specificity40 %, PPV 40 % and NPV 17 %.The dotted line (3A-B) represents80 % (standard cut-off level forbiomarkers) [6, 7]

846 Eur J Nucl Med Mol Imaging (2016) 43:839–851

in non-responders who also exhibited the greatest de-crease. The relevance of this observation is unknown buthighlights the limitation of using a marker of secretion asan indicator of tumor progression.

The strategy of integrating data from different sourcessuch as imaging, clinical and biological parameters intoindices or nomograms has previously been used to providea descriptive predictive tool for NETs. For example, anomogram composed of 15 variables (e.g., age, symptoms,5-HIAA, CgA, tumor size, invasion, metastasis, histology,Ki67 index and adopted therapy) has been useful in de-termining prognosis in small intestine NETs [41].Subsequently, a tumor size >4 cm combined with gradinghas been proposed to predict preoperative risk of lymphnode metastases in non-functioning pancreatic NETs [42].

More recently, an elevated quotient of gene expression(MORF4L2) and SUVmax has been reported as predictiveof disease status in NET patients undergoing 68Ga-SSA-PET [15]. In the current study, incorporating growth factorsignaling genes and those linked to metabolism activityprovided added dimensionality to the predictive capacityof the multianalyte algorithm for PRRT. It is essential,however, to continue the search for additional markers,likely genetic in origin, in order to build more accuratemodels and better characterize individuals with GEP-NETs.

Although PRRT is an effective therapy, the develop-ment of tools to facilitate timely and accurate identificationof therapeutic responses is needed. This is of particularrelevance in the management of NETs, and is crucial

Fig. 5 GF signalome, metabolome and grading in outcome prediction. a)Growth factor signaling gene expression (BGF signalome^) and genesinvolved in regulating cell metabolism (Bmetabolome^) weresignificantly higher in responders than non-responders at T=0 of CycleI. b) ROC analysis identified that each of these Bomes^ individually coulddifferentiate responders from non-responders. Metrics included growthsignaling (AUC=0.67±0.08, z-statistic=2.439, p=0.0147) and metabo-lome (AUC=0.7±0.09, z-statistic=2.262, p=0.0237). The BOme Index^

(a combination of GF signalome and metabolome) exhibited an AUC of0.74±0.08 (z-statistic=2.915, p=0.0036). This was more significant thaneach alone. c) Low-grade tumors (G1/G2; typical/atypical BP carcinoids)responded to therapy (77 %), while high-grade lesions (G3; BP undiffer-entiated) were associated with response (50 %). Grade alone, however,was not significant (p=0.12), d) ROC analysis of grade identified anAUC of 0.66 for treatment response prediction. NR=non-responder,R=responder

Eur J Nucl Med Mol Imaging (2016) 43:839–851 847

Table 5 Treatment Details

Patientno.

Sex PRRTprotocol

Primary Carcinoidsyndrome

Time sincediagnosis(months)

FDG Baselinestatus

Cum.activity(GBq)

RECISTresponse

Protocolterminationfor PD

Lengthof Fup(months)

11 M 1A Bronchial 0 10 Pos PD 20.4 SD No 22

16 M 1A Bronchial + 113 Pos PD 18.5 PD No 10

24 M 1A Bronchial 0 9 Pos PD 9.3 PD Yes 3

31 M 1A Bronchial 0 128 Pos PD 18.5 PD No 18

45 M 1A Bronchial + 5 Pos PD 18.9 SD No 16

49 M 1A Bronchial 0 81 NA PD 18.5 SD No 13

50 F 1A Bronchial 0 34 Pos SD 18.5 SD No 11

6 M 1A Pancreas + 37 NA PD 18.5 SD No 18

10 M 1A Pancreas 0 12 Pos SD 21.1 SD No 19

29 M 1A Pancreas 0 8 Pos SD 18.5 SD No 16

44 F 1A Pancreas 0 66 NA PD 17.9 SD No 13

52 M 1A Pancreas 0 25 Neg § 18.5 PR No 15

8 M 1A Jejunum 0 22 Pos SD 17.6 SD No 22

7 M 1A Ileum + 18 Pos PD 18.5 PD Yes 10

15 M 1A Ileum 0 49 Neg PD 18.5 SD No 18

20 F 1A Ileum + 110 Neg SD 18.5 PR No 16

21 F 1A Ileum + 176 Neg PD 18.5 SD No 19

35 M 1A Ileum + 14 Pos PD 18.5 SD No 18

42 M 1A Ileum + 75 Neg PD 8.3 PD Yes 3

51 M 1A Ileum 0 8 Pos PD 18.5 SD No 15

5 F 1A Unknown 0 22 Neg SD 18.5 SD No 17

13 F 1A Unknown + 9 NA SD 18.5 SD No 20

53 M 1B Bronchial 0 37 Pos PD 21.3 SD No 22

3 F 1B Bronchial 0 14 Pos SD 27.8 PD Yes 6

12 M 1B Bronchial 0 16 Pos PD 24.1 SD No 20

32 M 1B Bronchial 0 21 Pos PD 25.9 PR No 16

38 F 1B Bronchial 0 44 Neg SD 25 SD No 16

22 F 1B Pancreas 0 142 Neg PD 23.1 SD No 8

33 M 1B Pancreas 0 21 Pos PD 25.9 SD No 15

34 F 1B Pancreas 0 9 Pos § 23.1 PD No 17

40 M 1B Pancreas ZES 30 Neg PD 25.2 SD No 17

14 M 1B Ileum + 14 Pos PD 25.9 PD Yes 9

28 M 1B Ileum + 6 Pos SD 25.9 PR No 17

36 M 1B Ileum + 9 Pos PD 25.9 SD No 14

46 M 1B Ileum + 37 Pos SD 25.2 SD No 13

47 M 1B Ileum 0 3 Pos PD 25.9 SD No 14

1 M 1B Rectum 0 26 Neg PD 27.8 CR No 33

2 F 1B Unknown + 19 Neg PD 25.9 PR No 28

43 M 1B Unknown + 117 Pos PD 6.5 PD Yes 1

26 M 2 Bronchial 0 58 Neg PD 11.1 PD Yes 5

23 F 2 Gastric 0 55 Pos PD 14.8 PR No 10

9 M 2 Pancreas 0 51 Pos PD 15.7 SD No 17

25 M 2 Pancreas 0 52 Pos PD 14.8 SD No 16

27 M 2 Pancreas 0 59 Neg PD 14.8 SD No 19

37 M 2 Pancreas 0 213 Pos PD 14.8 PR No 17

48 F 2 Pancreas 0 135 Pos PD 14.8 SD No 15

4 F 2 Ileum + 111 Neg PD 14.8 SD No 25

848 Eur J Nucl Med Mol Imaging (2016) 43:839–851

when performing time/labor-intensive/expensive therapies(typically 6–12 months with the intended 177Lu-octreotatecumulative dosage) [28]. While imaging and imaging-based assessment of response (e.g., with RECIST criteria),in conjunction with biomarkers, is currently used to assesstreatment response, limitations include the difficulty incapturing volumetric modifications in slow-growing tu-mors, particularly for large-volume disease, or in definingtumor alterations using CT or MRI when structural chang-es, such as necrosis, hemorrhage or fibrosis, occur as aresult of treatment [43]. Various imaging protocols havebeen proposed to measure response, including the quanti-fication of SSR density in vivo and assessment of molec-ular tumor volume before and after PRRT [44].

Quantification criteria (SUVmax), however, are not yet pro-spectively validated [43]. Although the disease course andeffectiveness of PRRT can be followed by 177Lu-scans,early predictors of response to PRRT are rare [45, 46].The only significant factor is SSR density at baseline(OctreoScan; 68Ga-SSA-PET) [47]. An uptake atOctreoScan of grade 4 (higher than that of kidneys and/or spleen) is associated with an objective response in60 % of cases [32]. Intrinsic tumor and patient character-istics are also involved in response, as demonstrated byother predictors, including performance status, extent ofliver involvement and extra-hepatic metastases [3, 28].As a consequence, a personalized predictive assessmentrepresents a major unmet need.

Table 5 (continued)

Patientno.

Sex PRRTprotocol

Primary Carcinoidsyndrome

Time sincediagnosis(months)

FDG Baselinestatus

Cum.activity(GBq)

RECISTresponse

Protocolterminationfor PD

Lengthof Fup(months)

19 M 2 Ileum + 65 Neg PD 14.8 PR No 17

30 F 2 Ileum + 128 Pos SD 14.8 PD No 13

41 M 2 Ileum + 154 NA SD 17 PR No 17

18 F 2 Colonic 0 49 Pos PD 14.8 PD No 19

17 F 2 Rectum + 56 Neg PD 14.8 PD Yes 7

39 M 2 Unknown 0 83 Neg PD 14.8 PD No 17

54 M 2 Unknown 0 265 Neg PD 14.8 PD No 13

ZES, Zollinger-Ellison syndrome; Neg, negative; Pos, positive; NA, not assessable; ECOG, Eastern Cooperative Oncology Group performance status,PD, progressive disease; CR, complete response; PR, partial response; SD, stable disease; §, patient in response to chemotherapy; PFS, progression-freesurvival; Fup, follow up

Fig. 6 Derivation of a prediction quotient and clinical utility forpredicting responses. a) The combination of the BOme Index^ andtumor grading (WHO 2010 classification for GEP NETs; Travis for BPNETs) – the "Prediction Quotient^ had metrics of AUC: 0.90±0.06, z-statistic=6.39, p<0.0001, for predicting response in all tumors (n=53).This was significantly different from grading alone (difference betweenareas 0.23, z-statistic 2.25, p=0.024). b) Kaplan–Meier analysis ofprogression-free survival showed that the Response Quotient identified

that tumors that were Blow-grade/high-Ome index^ had an undefinedmedian survival. The Bhigh-grade/low-Ome index^ group had a medianprogression-free survival of 17 months. This was significant (log rank:26.8, p<0.0001). The hazard ratio was 53.3. c) The metrics for the Pre-diction Quotient were: accuracy: 94 %, sensitivity 79 %, specificity100 %, PPV 100 % and NPV 93 %. G(L)=low grade, G(H)=high grade,Ome(L)=low ome index, Ome(H)=high Ome index. The dotted linerepresents 80 % (accepted cut-off level for biomarkers) [6, 7]

Eur J Nucl Med Mol Imaging (2016) 43:839–851 849

Currently accepted biomarkers include CgA, despitereservations regarding its performance metrics and con-founding features such as renal insufficiency and protonpump inhibitor usage [48]. Moreover, many NET pa-tients (~30-50 %) do not have elevated CgA [49]. Inour series, 23 (43 %) of 54 patients had normal CgAlevels. Biochemical responses, however, are frequentlyobserved after PRRT [32], and a recent retrospectiveanalysis showed that a baseline CgA>600 ng/ml waspredictive of response [50]. This threshold was not pre-dictive in our prospective series. Tumor grading is alsoa well-recognized prognostic factor, although it is theexpression of a higher risk of tumor progression in thehistory of the patient rather than a predictor of the out-come of a specific treatment [30].

A limitation of this study is the inclusion of a cohortof advanced patients who, in some cases, had to discon-tinue treatment. These subjects, however, represent theBreal-world^ situation of advanced NET referral andtreatment. In this respect, a qPCR-based blood biomark-er test that defines the circulating Bfingerprint^ of aNET provides a novel strategy for assessing NET dis-ease. A circulating tumor MAAA can be easily acquired(simple blood draw) at multiple time points during pe-riods between sequential imaging assessment. Such in-formation can be combined with imaging results to pro-vide additional evidence regarding tumor behavior aswell as therapy response.

In conclusion, this study demonstrated that a circulat-ing NET gene transcript signature accurately correlatedwith treatment response and that a combination of genecluster analysis and grading—the predictive quotient—accurately predicted PRRT efficacy. Our study highlightstwo aspects of the assessment of PRRT efficacy withNET transcript analysis. One is the observation of thephenomena, namely, how the specific NET transcriptsignature behaves during and after PRRT, and how itcorrelates with standard morphologic and functional im-aging, and hence with the treatment response or out-come to therapy. This information is of clinical utilityin more precisely defining tumor growth status, thusconfirming the efficacy of therapy. The second aspectis the identification of specific gene clusters in pre-treatment blood that are able to predict the subsequentresponse to PRRT. Confirmation of these observationsin larger series will allow identification of likely non-responders and will better define at a molecular levelthe natural history of individual neuroendocrine tumors.

Compliance with ethical standards

Disclosure of potential conflicts of interest The authors declare thatthey have no conflict of interest.

Research involving human participants and/or animals All proce-dures performed in studies involving human participants were in accor-dance with the ethical standards of the institutional and/or national re-search committee and with the 1964 Declaration of Helsinki and its lateramendments or comparable ethical standards. This article does not con-tain any studies with animals performed by any of the authors.

Informed consent Informed consent was obtained from all individualparticipants included in the study.

References

1. Modlin IM, Oberg K, Chung DC, et al. Gastroenteropancreaticneuroendocrine tumours. Lancet Oncol. 2008;9:61–72.

2. van der Zwan WA, Bodei L, Mueller-Brand J, de Herder WW,Kvols LK, Kwekkeboom DJ. GEPNETs UPDATE: Radionuclidetherapy in neuroendocrine tumors. Eur J Endocrinol. 2015;172:R1–8.

3. Ezziddin S, Attassi M, Yong-Hing CJ, et al. Predictors of long-termoutcome in patients with well-differentiated gastroenteropancreaticneuroendocrine tumors after peptide receptor radionuclide therapywith 177Lu-octreotate. J Nucl Med. 2014;55:183–90.

4. Castano JP, Sundin A, Maecke HR, et al. Gastrointestinal neuroen-docrine tumors (NETs): new diagnostic and therapeutic challenges.Cancer Metastasis Rev. 2014;33(1):353–9.

5. Palmer C, Duan X, Hawley S, et al. Systematic evaluation of can-didate blood markers for detecting ovarian cancer. PLoS One.2008;3, e2633.

6. Frank R, Hargreaves R. Clinical biomarkers in drug discovery anddevelopment. Nat Rev Drug Discov. 2003;2:566–80.

7. Shapiro DE. The interpretation of diagnostic tests. Stat MethodsMed Res. 1999;8:113–34.

8. Oberg K. Diagnostic work-up of gastroenteropancreatic neuroen-docrine tumors. Clinics (Sao Paulo). 2012;67:109–12.

9. Khan MS, Kirkwood A, Tsigani T, et al. Circulating tumor cells asprognostic markers in neuroendocrine tumors. J Clin Oncol.2013;31:365–72.

10. Li SC, Essaghir A, Martijn C, et al. Global microRNA profiling ofwell-differentiated small intestinal neuroendocrine tumors. ModPathol. 2013;26:685–96.

11. Modlin IM, Drozdov I, Alaimo D, et al. A multianalyte PCR bloodtest outperforms single analyte ELISAs (chromogranin A,pancreastatin, neurokinin A) for neuroendocrine tumor detection.Endocr Relat Cancer. 2014;21:615–28.

12. Kidd M, Drozdov I, Modlin I. Blood and Tissue NET Gene ClusterAnalysis correlate, define Hallmarks and Predict Disease Status.Endocr Relat Cancer. 2015;22:561–75.

13. Modlin IM, Frilling A, Salem RR, Alaimo D, Drymousis P, WasanHS, et al. Blood measurement of neuroendocrine gene transcriptsdefines the effectiveness of operative resection and ablation strate-gies. Surgery. 2015. doi:10.1016/j.surg.2015.06.056.

14. Cwikla JB, Bodei L, Kolasinska-Cwikla A, Sankowski A, ModlinIM, KiddM. Circulating transcript analysis (NETest) in GEP-NETstreated with Somatostatin Analogs defines Therapy. J ClinEndocrinol Metab. 2015;100(11):E1437-45. doi:10.1210/jc.2015-2792.

15. Bodei L, Kidd M, Modlin IM, et al. Gene transcript analysis bloodvalues correlate with Ga-DOTA-somatostatin analog (SSA) PET/CT imaging in neuroendocrine tumors and can define disease status.Eur J Nucl Med Mol Imaging. 2015;42(9):1341–52.

850 Eur J Nucl Med Mol Imaging (2016) 43:839–851

16. Hanahan D, Weinberg RA. Hallmarks of cancer: the next genera-tion. Cell. 2011;144:646–74.

17. Oberg K, Modlin I, DeHerder W, et al. Biomarkers forNeuroendocrine Tumor Disease: A Delphic Consensus assessmentof Multianalytes, Genomics, Circulating Cells and Monoanalytes.Lancet Oncol. 2015;16, e435046.

18. Paganelli G, Sansovini M, Ambrosetti A, et al. 177 Lu-Dota-octreotate radionuclide therapy of advanced gastrointestinal neuro-endocrine tumors: results from a phase II study. Eur J Nucl MedMol Imaging. 2014;41(10):1845–51.

19. Sansovini M, Severi S, Ambrosetti A, et al. Treatment with theradiolabelled somatostatin analog Lu-DOTATATE for advancedpancreatic neuroendocrine tumors. Neuroendocrinology. 2013;97:347–54.

20. Severi S, Sansovini M, Ianniello A, Bodei L, Nicolini S, Ibrahim T,et al. Feasibility and utility of re-treatment with 177Lu-Dotatate inGEP-NENs relapsed after treatment with (90)Y-DOTATOC. Eur JNucl Med Mol Imaging. 2015;42(13):1955–63. doi:10.1007/s00259-015-3105-7.

21. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evalu-ation criteria in solid tumours: revised RECIST guideline (version1.1). Eur J Cancer. 2009;45:228–47.

22. Ezziddin S, Sabet A, Heinemann F, et al. Response and long-termcontrol of bone metastases after peptide receptor radionuclide ther-apy with (177)Lu-octreotate. J Nucl Med. 2011;52:1197–203.

23. Hanley JA, McNeil BJ. The meaning and use of the area under areceiver operating characteristic (ROC) curve. Radiology.1982;143:29–36.

24. Hanley JA, McNeil BJ. A method of comparing the areas underreceiver operating characteristic curves derived from the samecases. Radiology. 1983;148:839–43.

25. Bosman FT. WHO classification of tumours of the digestive sys-tem. 4th ed. Lyon: IARC Press; 2010.

26. TravisWD, Brambilla E, Muller-Hermelink HK, Harris CC. (Eds.):World Health OrganizationClassification of Tumours. Pathologyand genetics of tumours of the lung, pleura, thymus and heart.IARC Press: Lyon 2004.

27. Bodei L, Cremonesi M, Ferrari M, et al. Long-term evaluation ofrenal toxicity after peptide receptor radionuclide therapy with 90Y-DOTATOC and 177Lu-DOTATATE: the role of associated riskfactors. Eur J Nucl Med Mol Imaging. 2008;35:1847–56.

28. Kwekkeboom DJ, de Herder WW, Kam BL, et al. Treatment withthe radiolabeled somatostatin analog [177 Lu-DOTA 0,Tyr3]octreotate: toxicity, efficacy, and survival. J Clin Oncol.2008;26:2124–30.

29. Kwekkeboom DJ, de Herder WW, van Eijck CH, et al. Peptider e c e p t o r r a d i o n u c l i d e t h e r a p y i n p a t i e n t s w i t hgastroenteropancreatic neuroendocrine tumors. Semin Nucl Med.2010;40:78–88.

30. Rindi G, Petrone G, Inzani F. The 2010 WHO classification ofdigestive neuroendocrine neoplasms: a critical appraisal four yearsafter its introduction. Endocr Pathol. 2014;25:186–92.

31. Pelosi G, Papotti M, Rindi G, Scarpa A. Unraveling tumor gradingand genomic landscape in lung neuroendocrine tumors. EndocrPathol. 2014;25:151–64.

32. Kwekkeboom DJ, Kam BL, van Essen M, et al. Somatostatin-receptor-based imaging and therapy of gastroenteropancreatic neu-roendocrine tumors. Endocr Relat Cancer. 2010;17:R53–73.

33. Tannapfel A, Vomschloss S, Karhoff D, et al. BRAF genemutationsare rare events in gastroenteropancreatic neuroendocrine tumors.Am J Clin Pathol. 2005;123:256–60.

34. Perren A, Schmid S, Locher T, et al. BRAF and endocrine tumors:mutations are frequent in papillary thyroid carcinomas, rare in en-docrine tumors of the gastrointestinal tract and not detected in otherendocrine tumors. Endocr Relat Cancer. 2004;11:855–60.

35. Karhoff D, Sauer S, Schrader J, et al. Rap1/B-Raf signaling isactivated in neuroendocrine tumors of the digestive tract and Rafkinase inhibition constitutes a putative therapeutic target.Neuroendocrinology. 2007;85:45–53.

36. CookMR, Pinchot SN, Jaskula-Sztul R, Luo J, KunnimalaiyaanM,Chen H. Identification of a novel Raf-1 pathway activator that in-hibits gastrointestinal carcinoid cell growth. Mol Cancer Ther.2010;9:429–37.

37. Olsson AH, Yang BT, Hall E, et al. Decreased expression of genesinvolved in oxidative phosphorylation in human pancreatic isletsfrom patients with type 2 diabetes. Eur J Endocrinol. 2011;165:589–95.

38. Ruan Y, Cheng M, Ou Y, Oko R, van der Hoorn FA. Ornithinedecarboxylase antizyme Oaz3 modulates protein phosphatase ac-tivity. J Biol Chem. 2011;286:29417–27.

39. Hayflick SJ. Defective pantothenate metabolism and neurodegen-eration. Biochem Soc Trans. 2014;42:1063–8.

40. Valli A, Rodriguez M, Moutsianas L, et al. Hypoxia induces alipogenic cancer cell phenotype via HIF1alpha-dependent and -independent pathways. Oncotarget. 2015;6:1920–41.

41. Modlin IM, Gustafsson BI, Pavel M, Svejda B, Lawrence B,Kidd M. A nomogram to assess small-intestinal neuroendo-crine tumor ('carcinoid') survival. Neuroendocrinology.2010;92:143–57.

42. Partelli S, Gaujoux S, Boninsegna L, et al. Pattern and clinicalpredictors of lymph node involvement in nonfunctioning pancreaticneuroendocrine tumors (NF-PanNETs). JAMA Surg. 2013;148:932–9.

43 . Sund in A, Rocka l l A . The rapeu t i c moni to r ing o fgastroenteropancreatic neuroendocrine tumors: the challengesahead. Neuroendocrinology. 2012;96:261–71.

44. Baum RP, Kulkarni HR. THERANOSTICS: From MolecularImaging Using Ga-68 Labeled Tracers and PET/CT toPersonalized Radionuclide Therapy - The Bad Berka Experience.Theranostics. 2012;2:437–47.

45. Ezziddin S, Reichmann K, Yong-Hing C, et al. Early prediction oftumour response to PRRT. The sequential change of tumour-absorbed doses during treatment with 177Lu-octreotate.Nuklearmedizin. 2013;52:170–7.

46. Blaickner M, Baum RP. Relevance of PET for pretherapeutic pre-diction of doses in peptide receptor radionuclide therapy. PET Clin.2014;9:99–112.

47. Kratochwil C, Stefanova M, Mavriopoulou E, et al. SUV of[68Ga]DOTATOC-PET/CT Predicts Response Probability ofPRRT in Neuroendocrine Tumors. Mol Imaging Biol. 2015;17:313–8.

48. O'Toole D, Grossman A, Gross D, et al. ENETS ConsensusGuidelines for the Standards of Care in Neuroendocrine Tumors:biochemical markers. Neuroendocrinology. 2009;90:194–202.

49. Lindholm DP, Oberg K. Biomarkers and molecular imaging ingastroenteropancreatic neuroendocrine tumors. Horm Metab Res.2011;43:832–7.

50. Sabet A, Dautzenberg K, Haslerud T, et al. Specific efficacy ofpeptide receptor radionuclide therapy with (177)Lu-octreotate inadvanced neuroendocrine tumours of the small intestine. Eur JNucl Med Mol Imaging. 2015;42:1238–46.

Eur J Nucl Med Mol Imaging (2016) 43:839–851 851